fix: set SSE reconnect delay to 100ms

[difhel]

Summary

Send retry: 100\n\n instead of the initial blank line in both v1 and v3 SSE handlers, using the existing immediate flush. Update the v3 write-error message to describe the new initial frame.

This tells native EventSource clients to use a 100 ms reconnection interval. Without an explicit SSE retry field, WebKit defaults to 3000 ms, which can add a noticeable delay when a backgrounded wallet reconnects.

[github-actions]

📊 Performance Metrics — PR vs main

memory — ✅

Metric	main	PR	Δ
CPU	—s	0.72s	—
Goroutines	—	11	—
Threads	—	11	—
Heap	—MB	15.02MB	—
RAM	—MB	46.56MB	—
Total Alloc	—MB	37.58MB	—
Allocs	—	273487	—
GC Cycles	—	11	—
GC Avg	—ms	0.41ms	—
FDs	—	9	—

postgres — ✅

Metric	main	PR	Δ
CPU	—s	0.87s	—
Goroutines	—	13	—
Threads	—	11	—
Heap	—MB	22.06MB	—
RAM	—MB	54.16MB	—
Total Alloc	—MB	40.90MB	—
Allocs	—	296189	—
GC Cycles	—	10	—
GC Avg	—ms	0.43ms	—
FDs	—	35	—

cluster-valkey — ✅

Metric	main	PR	Δ
CPU	—s	0.66s	—
Goroutines	—	14	—
Threads	—	12	—
Heap	—MB	28.00MB	—
RAM	—MB	67.70MB	—
Total Alloc	—MB	37.34MB	—
Allocs	—	258511	—
GC Cycles	—	8	—
GC Avg	—ms	0.59ms	—
FDs	—	71	—